Estimating forest canopy bulk density using sixindirect methods

Robert E. Keane, Elizabeth D. Reinhardt, Joe Scott, Kathy Gray, andJames Reardon

Abstract: Canopy bulk density (CBD) is an important crown characteristic needed to predict crown fire spread, yet itis difficult to measure in the field. Presented here is a comprehensive research effort to evaluate six indirect samplingtechniques for estimating CBD. As reference data, detailed crown fuel biomass measurements were taken on each treewithin fixed-area plots located in five important conifers types in the western United States, using destructive samplingfollowing a series of four sampling stages to measure the vertical and horizontal distribution of canopy biomass. Thesix ground-based indirect measurement techniques used these instruments: LI-COR LAI-2000, AccuPAR ceptometer,CID digital plant canopy imager, hemispherical photography, spherical densiometer, and point sampling. These tech-niques were compared with four aggregations of crown biomass to compute CBD: foliage only, foliage and smallbranchwood, foliage and all branchwood (no stems), and all canopy biomass components. Most techniques had the bestperformance when all canopy biomass components except stems were used. Performance dropped only slightly whenthe foliage and small branchwood canopy biomass aggregation (best approximates fuels available for crown fires) wasemployed. The LAI-2000, hemispherical photography, and CID plant canopy imager performed best. Regression equa-tions that predict CBD from gap fraction are presented for all six techniques.

Résumé : La densité apparente de la canopée (DAC) est une caractéristique importante de la cime qui est nécessairepour prédire la propagation d’un feu de cime mais qui est cependant très difficile à mesurer sur le terrain. Les auteursprésentent ici un travail de recherche exhaustif dont le but était d’évaluer six techniques indirectes d’échantillonnagepour estimer la DAC. Les données de référence proviennent de mesures détaillées de la biomasse des combustiblesdans la cime prises sur chaque arbre dans des placettes à superficie fixe situées dans cinq types importants de forêtrésineuse de l’ouest des États-Unis en utilisant une approche destructrice après avoir procédé à une séried’échantillonnages en quatre étapes pour mesurer la distribution verticale et horizontale de la biomasse dans la ca-nopée. Les six techniques indirectes de mesure sur le terrain comprenaient les instruments suivants : le LAI-2000 deLI-COR, le ceptomètre d’AccuPAR, l’imageur digital du couvert végétal CID, la photographie hémisphérique, le den-siomètre sphérique et l’échantillonnage par point. Ces techniques ont été comparées à quatre regroupements de la bio-masse de la cime pour calculer la DAC : feuillage seulement, feuillage et petites branches, feuillage et toutes lesbranches (sans la tige) et toutes les composantes de la biomasse de la cime. La plupart des techniques offraient lameilleure performance lorsque toutes les composantes de la biomasse de la canopée, à l’exception de la tige, étaientutilisées. La performance diminuait juste un peu avec l’utilisation du regroupement de la biomasse de la canopée com-posé du feuillage et des petites branches (regroupement qui fournit la meilleure approximation des combustibles dispo-nibles pour les feux de cime). Le LAI-2000, la photographie hémisphérique et l’imageur digital du couvert végétal CIDont eu la meilleure performance. Des équations de régression qui prédisent la DAC à partir de la proportiond’ouvertures sont présentées pour les six techniques.

[Traduit par la Rédaction] Keane et al. 739

Introduction

Successful fire-suppression programs in the western UnitedStates and Canada over the last 70 years have reduced fire

occurrence in many fire-prone forests, resulting in excessivebuildups of fuels, especially tree crowns, which has in turnincreased the probabilities and intensities of future severecrown fires (Mutch 1994; Ferry et al. 1995; Kolb et al. 1998;Keane et al. 2002). Crown fires have become especiallycommon in dry, low-elevation forests that, prior to Europeansettlement (ca. 1900), frequently experienced nonlethal sur-face fires or mixed-severity fires that rarely burned overstorytree crowns and were somewhat easy to control (Kolb 1998;Arno et al. 2000). Therefore, it is very important that theforest canopy characteristics be quantified to assess crownfire hazard, prioritize treatment areas, and design treatmentsto reduce crown fire potential.

One canopy characteristic important for crown fire propa-gation is canopy bulk density (CBD), defined as the mass

Can. J. For. Res. 35: 724–739 (2005) doi: 10.1139/X04-213 © 2005 NRC Canada

724

Received 13 August 2004. Accepted 25 November 2004.Published on the NRC Research Press Web site athttp://cjfr.nrc.ca on 8 April 2005.

R.E. Keane,1 E.D. Reinhardt, K. Gray, and J. Reardon.USDA Forest Service, Rocky Mountain Research Station, FireSciences Laboratory, P.O. Box 8089, Missoula, MT 59807,USA.J. Scott. Systems for Environmental Management, P.O.Box 8868, Missoula, MT 59801, USA.

1Corresponding author (e-mail: rkeane@fs.fed.us).

per unit volume of canopy biomass that would burn in acrown fire (primarily foliage and twigs less than 3 mm in di-ameter) (see Fig. 1) (van Wagner 1977; Alexander 1988;Finney 1998b). A number of fire behavior and effects modelsrequire estimates of CBD and several other canopy fuel char-acteristics (namely canopy base height, stand height, andcanopy cover) to accurately simulate crown fires, includingthe models FIRETEC (Linn 1997) and NEXUS (Scott 1999).The fire growth model FARSITE (Finney 1998), currentlyused by many fire management agencies, requires CBD andthe other canopy characteristics to be mapped across thesimulation landscape at high spatial resolution to predictcrown fire behavior. Unlike the other canopy characteristics,standardized field methods do not exist for sampling CBD.Fire managers urgently need an efficient field sampling tech-nique to estimate CBD for simulating crown fire behavior tofacilitate design, implementation, and monitoring of canopyfuel treatment projects.

We evaluate six indirect techniques for their ability toestimate forest CBD for fire management applications. Forreference data, we measured crown biomass on all treeswithin fixed-area plots using destructive sampling followinga series of cutting treatments on five study sites representingcommon western United States conifer types. We evaluatedsix ground-based, optical, indirect measurement methods forestimating CBD. These methods use equipment for estimat-ing leaf area index (LAI) and canopy gap fraction. Resultsfrom CBD estimates are compared across the six indirecttechniques, and regression equations are developed so thatthese methods can be used in the field. Recommendations ontheir use are discussed.

Background

Current methodsCanopy bulk density (CBD) is difficult to measure be-

cause it requires detailed knowledge of the vertical distribu-

tion of crown biomass (Alexander 1988) (Fig. 1). Directmethods of destructively sampling tree biomass by verticalcanopy layers are expensive and time intensive (Gary 1978;Scott and Reinhardt 2002). We found only two studies thatmeasured vertical canopy fuel distribution. Sando and Wick(1972) computed CBD for 0.3-m canopy layers across alodgepole pine forest vertical profile, and Gary (1976, 1978)intensively sampled thinned and unthinned 80-year-old lodge-pole pine stands to determine crown biomass distributionand structure.

The most popular method for estimating CBD uses mea-surements of tree diameter, height, and crown base heightfor all trees in a stand to calculate crown biomass distribu-tion from allometric crown biomass equations (Keane et al.1998, 2000). Reinhardt and Crookston (2003) used the Sandoand Wick (1972) approach in combination with Brown’s(1978) crown equations to estimate canopy bulk density fromstand inventory data (tree density, species, diameter at breastheight (dbh; diameter at 1.37 m), height, crown base height).However, crown biomass equations are not available for alltree species, size classes, and stand conditions, so this methodis impractical for many forested ecosystems. Johnson et al.(1989) developed crown fuel component biomass equationsfor lodgepole pine and white spruce using tree height andcrown width but did not predict the vertical distribution ofcanopy fuels. The allometric approach for predicting CBDhas not been previously validated and is restricted to thoseforest types with appropriate biomass equations.

Because the vertical distribution of CBD is highly vari-able in a stand, average values across the canopy profile maynot adequately represent the fuel conditions required for crownfire propagation (Fig. 1). Crown fire spread may dependonly on a few dense canopy layers with high CBD. Verticalcanopy fuel characteristics are associated with species com-position and stand structure where shade-tolerant species tendto occupy the lower canopy and tend to have higher propor-tions of flammable foliage and fine fuel than shade-intolerant,

© 2005 NRC Canada

Keane et al. 725

Fig. 1. Crown bulk density profile for a hypothetical stand. The crown bulk density averaged across the entire profile is 0.2 kg·m–3,but one canopy layer at about 5 m high has a bulk density of 0.4 kg·m–3.

early-seral species (Brown 1978; Roberts and Long 1992;Keane et al. 2002). Therefore, any estimate of CBD mustdescribe those canopy layers that account for crown firespread.

TheoryGap fraction is defined as the proportion of the sky visible

when viewed from below the canopy, and it has been suc-cessfully used to estimate leaf area index (LAI) in croplands,shrublands, and forests (Welles 1990; Chason et al. 1991;Chen et al. 1993; Martens et al. 1993; Dufrene and Breda1995; Nilson 1999). LAI is linearly related to foliar crownbiomass by specific leaf area (unit leaf area per dry mass), ifit is assumed that biomass of branches supporting foliage isproportional to needle biomass (Brown 1978). Therefore, thegap fraction of a canopy should be related to the amount ofcrown biomass and CBD based on the following theory.

Each ray of radiation passing through the vegetation canopyhas a chance that it will be intercepted by foliage or support-ive tissue, and this chance is proportional to path length, fo-liage density (FD, foliage area per canopy volume, m2·m–3),and foliage orientation (G) (Welles and Norman 1991; Nilson1999). In a horizontally homogeneous canopy of infinitelythin, planar leaves, the transmittance or gap probability, T,can be represented by the equation

[1] T G S( , ) [ ( ) ( , ) ( , )]θ φ θ φ θ φ= −e FD

where T( , )θ φ represents the gap fraction (the probability thata ray of light will not be intercepted at zenith angle θ and az-imuth angle φ), S( , )θ φ is the path length of the ray throughthe canopy, and G( , )θ φ is the foliage orientation (assumedto be one for infinitely thin planar leaves for simplicity)(Welles and Norman 1991; Stenberg et al. 1994). A solutionfor foliage density comes from Miller (1967) as

[2] FD d= −∫20

2ln ( )

( )sin

/T

Sθ

θθ θ

π

This assumes random distribution of canopy elements (fo-liage, stems, branches) in all directions, so the azimuth angleφ is dropped (Welles and Norman 1991). Foliage density isdirectly related to CBD by canopy height (z, m) and specificleaf area constants (SLA, m2·kg–1) which are species specific(Pierce et al. 1994), given the equation CBD = FD × SLAfor each canopy layer across canopy height (z). Then, pathlength S is computed from canopy height and zenith angle θin the equation S z( ) /cosθ θ= . Combining these relationshipsand summing across n zenith angles θ, we get the equation

[3] CBDSLA

= −

=∑2

1

ln sin cosTz

i i i

i

n θ θ

where Ti is the gap fraction measured at zenith angle θi andz is canopy height. Estimation of CBD from eq. 3 is highlyinfluenced by many factors, including foliage orientation (rep-resented by the function G(θ,φ)) (Stenberg 1996a; Kuchariket al. 1998), foliage clumping (Fassnacht et al. 1994; Stenberg1996a), SLA changes over canopy depth, and the measure-ment error of the gap fraction (Martens et al. 1993). Thisequation indicates that CBD can be estimated from threevariables: gap fraction, stand height, and specific leaf area.

There are three common methods for estimating gap frac-tion. It can be optically estimated directly from digital hemi-spherical images taken below the canopy looking up or abovethe canopy looking down (Chen et al. 1993; Frazer et al.1997; Nilson et al. 1999). Gap fraction can also be estimatedas a proportion of foliage intercepts using active remotelysensed imagery, such as Lidar (Magnussen and Boudewyn1998), or sampling techniques that compute gap fractionfrom a proportion of foliage intercepts (Clark and Seyfried2001). Finally, gap fraction can be estimated indirectly asthe fraction of radiation reaching the forest floor (i.e., below-canopy radiation divided by above-canopy radiation) mea-sured in specific wavelength ranges (Nel and Wessman 1993).

Indirect measurement techniquesThis study evaluated six common indirect methods to esti-

mate CBD. All methods use canopy gap fraction to computea canopy characteristic, such as leaf area index or canopycover, from readily available instruments and sampling tech-niques. We selected these methods based on their ability to beused in extensive fuel sampling efforts and their success inmeasuring other canopy characteristics. The instruments were theLI-COR LAI-2000 and AccuPAR ceptometer, which esti-mate gap fraction by the fraction of below- and above-canopyradiation; the CID digital plant canopy imager (CIDpci), hemi-spherical photographs, and spherical densiometer, which com-pute gap fraction from digital images; and point samplingusing vertical line intercept protocols (Table 1).

Perhaps the most commonly used instrument to indirectlymeasure LAI is the LAI2000 plant canopy analyzer(LAI2000) (LI-COR 1992). The LAI2000 estimates gap frac-tion as the fraction of radiation transmitted through the can-opy at five zenith angles (7°, 23°, 38°, 53°, and 68° fromvertical) measured with five concentric lenses on a handheldwand (Welles and Norman 1991). Stenberg et al. (1994) notethat the instrument does not sense leaf area, but rather leafand shoot silhouette area, which includes stems, branches,reproductive organs, and arboreal lichens and mosses, whichmay be a desirable trait when estimating CBD. As a result,the LAI2000 tends to underestimate LAI in stands with highleaf area (>5.0 LAI) because supporting tissue may blockclumped foliage (White et al. 1998). Correction factors toadjust LAI2000 measurements have been developed for manyecosystems (Gower and Norman 1991; Smith et al. 1993;Stenberg et al. 1994; Smolander and Stenberg 1996; Whiteet al. 1998), but Sampson and Allen (1995) note that species-specific correction factors will not necessarily correct bias inLAI2000 measurements because of high variability in the spa-tial distribution of species and crown structure. Ideal conditionsfor using the LAI2000 are uniformly cloudy days or neardawn or dusk (LI-COR 1992; Peper and McPherson 1998).

The AccuPAR Sunfleck Ceptometer (ceptometer) is a lin-ear ceptometer consisting of a wand containing 80 sensors at1-cm intervals that measure photosynthetically active radia-tion in the 400–700 nm wavelength range (Decagon 1987).The sensor arrangement attempts to mitigate effects of foliarclumping by detection of sunflecks, and a microprocessorscans the 80 sensors and calculates the arithmetic radiationaverage (Peper and McPherson 1998). Gap fraction and LAIcan be computed using the Campbell and Norman (1989)method or by the division of measurements taken inside and

© 2005 NRC Canada

726 Can. J. For. Res. Vol. 35, 2005

outside the canopy. The main problem with the ceptometer,as with most quantum sensors, is that the light environmentunder and above the forest canopy is constantly changing soit is difficult to synchronize clear-sky and below-canopyradiation measurements and it is difficult to obtain meaning-ful measurement replicates in the sample stand. Clear skyand low zenith angles are needed for the best ceptometerreadings.

Hemispherical photography (hemiphoto) is gaining popu-larity for the computation of leaf area and the exploration ofcanopy light dynamics (Neumann et al. 1989; Englund et al.2000; Frazer et al. 2001). Hemispherical photographs aretaken directly upwards from the ground to the canopy on alevel tripod using a fish-eye lens and high-resolution film ordigital photography (Frazer et al. 1997). Digital images areimported into a variety of analysis software to compute vari-ous vegetation indices and light regimes (Smith and Somers1991; Inouye 2000; Frazer et al. 2000). Lens optics, expo-sure settings, scanner resolution, and nonlevel tripod can sig-nificantly affect digital photo quality (Walter and Himmler1996). Hemiphotos can be taken under a wide variety oflight conditions but Frazer et al. (2001) found the best re-sults in uniformly overcast skies.

The CID CI-100 digital plant canopy imager (CIDpci)(CID, Inc., Vancouver, Washington, USA) takes a high-resolution digital image from a self-leveling camera mountedat the end of a probe, very much like the hemiphoto tech-nique. The image is stored on the hard disk of a field com-puter for future processing. The CIDpci software allowsvarious modifications, transformations, and masking of thedigital image to enhance calculation of gap fraction, includ-ing dividing the image into a number of zenith and azimuthangle partitions. Gap fraction is computed from the fractionof sky visible in each partition. This instrument can be usedunder a wide variety of weather conditions from sunny tocloudy.

The spherical densiometer (densiometer) is commonly usedto estimate crown closure for many forestry applications be-cause it is inexpensive and relatively easy to use. It consistsof a convex mirror with a grid of 24 squares etched into theglass, with each square containing four points. The densiometeris held level underneath the canopy, and canopy closure is

computed as a tally of those points out of 96 that are cov-ered by foliage (gap fraction = 1 – canopy cover). This devicehas had limited success because it is difficult to hold the in-strument level and still during the tally process. It can beused under most weather conditions with minimal training.

Point intercept sampling techniques (point) have been usedto estimate canopy characteristics, primarily canopy closure,for many forest and range communities (Battles et al. 1996;Groeneveld 1997). Most involve installing a number of sam-pling points in a grid and then recording whether a line ex-tending directly down or up from the sample point intersectsplant material (foliage or branches) (Jurik et al. 1986). Weused a device that contains a prism with cross-hairs that al-lows an upward-looking view of the canopy while keepingthe head level. The major limitation of point sampling is thatit takes a large number of points to accurately describe gapfraction, and these points should be along a grid or randomcircuit to minimize bias.

Several studies have compared these methods and instru-ments for estimating LAI or some other crown characteris-tic. Dufrene and Breda (1995) evaluated the LAI2000, aDemon portable light sensor, and a net radiometer to com-pute LAI as calculated from litter-trap data and found theLAI2000 provided the best estimates. Chason et al. (1991)compared the LAI2000 and a direct beam sensor on a tramsystem with litterfall estimates of LAI and found that theLAI2000 performed well when measurements were scaled tocorrect for clumping. Martens et al. (1993) compared theceptometer, another line quantum sensor, the LAI2000, andhemispherical photographs with direct biomass measurementand found the LAI2000 was easiest to use and performed thebest in an orchard setting but was less reliable in a forest.The ceptometer tended to overestimate LAI, contrary to re-sults from Pierce and Running (1988). An extensive evalua-tion study was done by Peper and McPherson (1998) wherethey compared a ceptometer, LAI2000, CIDpci, allometricequations, and hemispherical photography to estimate LAI.They found all methods appeared to underestimate LAI andnone produced satisfactory estimates, but overall hemispher-ical photography seemed to perform the best. Similar resultswere found by Fassnacht et al. (1994) when they compared aceptometer, LAI2000, and DEMON line sensor. The present

© 2005 NRC Canada

Keane et al. 727

Instrumenta CompanyEstimatedcost

Ease ofuse Method

LAI2000 LI-COR $4200 Moderate Gap fraction estimated from canopy transmittance forfive zenith angles

Ceptometer AccuPAR $2900 Easy Gap fraction estimated at various azimuth anglesDigital plant canopy imager

(CIDpci)CID, Inc., Vancouver,

Washington, USA$4200 Moderate Gap fraction estimated from digital imagery taken

with a high-resolution camera on a wand con-nected to a computer that processes the imagery

Hemispherical photography(hemiphoto)

Several companiesprovide instrumentsand software

$600–$2500 Moderate Gap fraction estimated from upward-pointing digitalphotography analyzed using special software

Spherical densiometer(densiometer)

Available at manyforestry suppliers

$100 Easy Gap fraction estimated by counting empty grid pixelsin an upward-pointing convex mirror

Point sampling (point) Available at manyforestry suppliers

$40 Easy Gap fraction estimated by the number of vertical linetransects that do not intercept foliage

aItalicized names are the labels used to represent the indirect techniques in this paper.

Table 1. Description of the six indirect measurement techniques used to estimate canopy characteristics in this study.

study differs from other comparison studies in that we aretesting instruments for estimation of CBD, not LAI. BecauseCBD includes the small supporting branches, these indirectmethods may estimate CBD more accurately than LAI.

Materials and methods

General descriptionFive sites representing major conifer types in the western

United States were selected for this study (Table 2). Thesesites historically experienced nonlethal surface or mixed-severity fire regimes but have recently become susceptible tocrown fire because of a buildup of overstory and understorycrown fuels, resulting from the lack of fires. Indirect mea-surements were taken at 25 points within a fixed-area plotlocated in a relatively homogeneous portion of the site basedon crown fuel characteristics. The plot size was varied bystudy site to ensure adequate representation of stand condi-tions as judged from plot inventories of tree density data(Table 2).

To simulate a wide range of crown conditions, we re-moved approximately 25% of tree basal area from each plotin three sequential sampling stages to create four or five(Ninemile, Blodgett, and Tenderfoot sites had an extra treat-ment) different stand structures for each of the five samplesites, ultimately resulting in 23 observations (5 sites × 4 or5 sampling stages). Each tree with dbh > 1 in. (1 in. =2.54 cm) was tagged, and its dbh was recorded along withspecies and other descriptive attributes. An initial set of indi-rect measurements was taken once the grid was installed.Then, the smallest trees were destructively sampled untilapproximately 25% of the basal area was removed, and an-other set of indirect measurements was taken. This was re-peated four times (25%, 50%, 75%, and 100% basal arearemoval), always removing the smallest remaining trees. Thisminimized spatial clumping and ensured removal proceededfrom the understory to overstory. Care was taken to protectstanding trees and branches, especially saplings, during cut-ting. Trees outside the plot boundaries, but within one treeheight of the plot edge, were also cut at each sampling stageusing the same criteria for removal. These trees were notsampled for crown characteristics; however, they were limbedand the branches were scattered so that none of the cut treewas visible to the indirect instruments.

Before each tree was cut, a complete inventory of all deadand live branches was taken at 1 m height intervals by climb-ing and pruning each tree branch. Then, 10% of the cutbranches were systematically sampled to estimate the wetmass of the following crown fuel components: (1) dead andlive branchwood in the diameter classes of 0–3, 3–6, 6–10,10–25, 25–50, and >50 mm; (2) live and dead foliage;(3) cones; and (4) lichen and moss. This time-intensive taskinvolved stripping all dead and live foliage from the sub-sample branch by hand and weighing the foliage on a porta-ble scale. Lichens, mosses, and cones were also separatedfrom the branch and weighed. Branchwood was cut into thesize classes by sliding a fixed-width gauge up a branch seg-ment and then cutting the branch when a threshold diameterclass was reached. A small subsample from each crown fuelcomponent pile was collected and placed in an oven at 80 °Cfor 2 days to determine moisture content. Green mass for all

© 2005 NRC Canada

728 Can. J. For. Res. Vol. 35, 2005

Stu

dyar

eaC

over

type

aL

and

man

agem

ent

agen

cyE

leva

tion

(m)

Asp

ectb

Bas

alar

ea(m

2 ·ha–1

)

Ove

rsto

ryde

nsit

y(t

rees

·ha–1

)c

Und

erst

ory

dens

ity

(tre

es·h

a–1)

Avg

.ov

erst

ory

heig

ht(m

)

Plo

tsi

ze(h

a)

Nin

emil

eP

onde

rosa

pine

/D

ougl

as-f

irL

olo

Nat

iona

lF

ores

t,M

onta

na10

50N

NE

30.5

481

2514

220.

07S

alm

onD

ougl

as-f

ir/

lodg

epol

epi

neS

alm

on-C

hall

isN

atio

nal

For

est,

Idah

o23

00S

E37

.711

7810

3317

0.03

Ten

derf

oot

Lod

gepo

lepi

neL

ewis

and

Cla

rkN

atio

nal

For

est,

Mon

tana

2290

NE

42.7

1145

2300

190.

03F

lags

taff

Pon

dero

sapi

neC

ocon

ino

Nat

iona

lF

ores

t,C

entr

alA

rizo

na23

08S

6920

6763

315

0.03

Blo

dget

tS

ierr

aN

evad

am

ixed

coni

fer

Blo

dget

tF

ores

tR

esea

rch

Sta

tion

,C

alif

orni

a13

00N

NE

46.8

382

142

340.

07a C

over

-typ

esp

ecie

s:po

nder

osa

pine

(Pin

uspo

nder

osa

var.

pond

eros

a),

Dou

glas

-fir

(Pse

udot

suga

men

zies

ii),

lodg

epol

epi

ne(P

inus

cont

orta

var.

cont

orta

).b N

,no

rth;

S,so

uth;

E,

east

;W

,w

est.

c Ove

rsto

ryar

eth

ose

tree

sw

ithdi

amet

erat

brea

sthe

ight

>10

cm.

Tab

le2.

Gen

eral

desc

ript

ion

ofth

est

udy

area

sin

clud

edin

this

stud

y.

crown fuel components was converted to dry mass using themoisture content for that sample (Deusen and Baldwin 1993).

Several branches from each species of tree on the sitewere taken back to the laboratory for computation of spe-cific leaf area (SLA). Branches were clamped onto a stand,and a digital photograph was taken directly down against aneutral background. This image was imported into softwareto compute the projected area of the branch including nee-dles and branchwood. The needles were then stripped off thebranch by hand, weighed, counted, and then mounted onwhite paper. Each paper was scanned using a scanner andassociated software to compute projected needle area andneedle length. One needle from each paper was cut in half,and the long and short axes of the cross-section were mea-sured. Total needle area was then computed for each needlefrom the length, width, and height using the cross-sectionalshape geometry of a half-sphere, rectangle, or triangle, de-pending on the species. The branch projected leaf area, nee-dle projected leaf area, and total needle area was divided bythe needle mass to calculate three measures of SLA.

Biomass estimates (kg) by crown fuel component werecomputed for each branch using regression equations devel-oped from the subsampled branches by tree species (Alemdag1980; Monserud and Marshall 1999). The dry mass of eachcrown fuel component was predicted from measured branchcharacteristics (e.g., length, width, foliar ratio, mass) on thesubsampled branches. Regression equations were applied tothe unsampled branches to compute crown biomass by fuelcomponent.

These crown fuel data were summarized to bulk densitymeasures by 1-m intervals for the entire stand profile, andthey were used as dependent variables in the indirect mea-surement evaluation. Gap fractions estimated from the sixtechniques were related to measured CBD using regressionanalysis. Detailed descriptions of crown fuel sampling meth-ods and developed crown biomass equations are presented inScott and Reinhardt (2002), so only indirect measurementtechniques and their analyses are presented here.

Indirect measurementsA 5 × 5 grid of 25 points was established inside plot

boundaries with each point 5 m apart. The grid was orientedup and down the slope, and if the slope was flat, it was ori-ented north and south. Each point was marked with a 15-cmnail attached to yellow flagging. A crew of four extensivelytrained technicians took measurements for each of the six in-direct instrument techniques at each point at waist level (1 mabove ground).

The LAI2000 wand was positioned directly over the sam-pling point and above the undergrowth. Most measurementswere taken at dawn or dusk or if the sky was heavily over-cast. One calibration reading was taken in the open outsidethe stand before and after the 25 inside-stand measurements.All readings for the 25 points were stored in the instrumentand downloaded to a computer for postprocessing analysisthat included eliminating spurious readings (transmittance >1.0 for all zenith angles).

Radiation flux was measured with the ceptometer at eachsample point as an average of four measurements taken bypointing the level wand in the four cardinal directions. Eachmeasurement was taken as an average of PAR flux density

(W·m–1) across all sensors on the wand. All fourmeasurements for each point, taken as close to solar noon aspossible, were stored in the instrument and downloaded atday’s end. Four calibration readings outside the stand in anopening were taken directly before and after the 25 below-canopy readings.

A high-resolution digital image was taken at each of the25 sample points with the CIDpci and hemispherical photog-raphy (hemiphoto) and stored on a laptop computer. TheCIDpci wand was pointed downslope or north if the slopewas flat. CIDpci images were analyzed with the accompany-ing CID software. Hemispherical photographs were takenwith a Nikon 990/995 digital camera with a Coolpix 900fish-eye lens mounted on a self-leveling tripod positionedabout 1 m directly above the sample point. Exposure and f-stop settings depended on light conditions and were selectedby the camera. High-resolution digital photographs (1024 ×1024 pixels) were imported into the Hemiview software(Delta T devices, Inc., Cambridge, UK; http://www.delta-t.co.uk) for computation of gap fraction.

The densiometer and point sampling tube were held levelby hand directly above each sample point. The number ofpoints in the densiometer grid that contained clear sky (i.e.,no canopy material) was counted on the concave mirror. Thepoint sample was taken by holding the sampling tube directlyvertical and sighting through the tube and recording if thecross-hairs did not intercept canopy material. Both estimateswere recorded on a plot form.

Data analysisWe used three canopy metrics and four aggregations of

crown biomass components to evaluate the performance ofthe indirect techniques across diverse representations of theforest canopy (Table 3). The canopy metrics are (1) canopyloading (CL), canopy biomass per unit area (kg·m–2); (2) av-erage CBD (CBDave), computed as an average of CBD acrossall 1-m canopy layers (kg·m–3); and (3) maximum CBD(CBDmax), the maximum CBD value (kg·m–3) across allcanopy layers.

These metrics were selected to represent different expres-sions of canopy characteristics that might be useful to firemodeling and land management and that might be detectedusing the indirect techniques. Each canopy metric was com-puted using four different aggregations of canopy biomassbased on groupings of the crown components (Table 3):(1) foliage only, (2) foliage, moss and lichen, and all branch-wood less than 3 mm diameter (this is the material assumedto be available to a crown fire); (3) all foliage and branch-wood excluding tree stems greater than 75 mm diameter;and (4) all canopy material: foliage, branchwood, and stems.

The canopy metrics were computed by summing the ap-propriate crown biomass components for canopy biomassaggregation across all trees for each 1 m height layer, andthen dividing by plot area. This yields 12 unique canopymeasures. Large stems and branches are not considered can-opy fuel because they are rarely consumed in a crown fire,but they were included in this analysis because they were de-tected by all indirect techniques and their presence cannot beremoved from most of the indirect measurements.

Gap fractions estimated from the six indirect techniques ateach of the four stand conditions (uncut, 25% removal, 50%

© 2005 NRC Canada

Keane et al. 729

removal, 75% removal) on each plot were related to the12 canopy measures using correlation and regression tech-niques. For the LAI2000, CIDpci, and hemiphoto methods,gap fractions were computed using five schemes that useddifferent sets of zenith angles and averaging procedures (Ta-ble 3). First, gap fraction was averaged across all five zenithangle ranges (7°, 23°, 38°, 53°, and 68° from vertical forLAI2000 and 9°, 27°, 45°, 63°, and 81° for hemiphoto andCIDpci), then across the top three angles, and lastly, fromonly the topmost zenith angle. In addition, gap fraction wascomputed as a weighted average across the five and threezenith angles using a transformation from eq. 3 where theweight (w) is

[5] wT

=−∑ (ln ) sin cosθ θ

2

where θ is the midpoint of the range of zenith angles and Tis gap fraction (Table 3). For the ceptometer, the average ofthe four below-canopy ceptometer measurements for eachsample point was divided by the calibration or outside-canopymeasurements to compute gap fraction. Gap fraction for thedensiometer was approximated by the number of clear-skypoints divided by the total number of points (94) on the con-cave mirror averaged across all sample points. For the pointtechnique, gap fraction was computed for the entire stand as

the number of clear-sky sample points divided by 25 (totalnumber sample grid points).

To evaluate indirect method performance, we conductedan extensive correlation analysis where the Pearson’s corre-lation coefficient (r) was used to evaluate the ability of eachindirect technique to consistently estimate the canopy repre-sentations across the five study sites and four or five samplestages. We performed a second correlation analysis on theLAI2000, hemiphoto, and CIDpci measurements to evaluatethe five schemes of computing gap fraction for predictingthe canopy characteristics (see Table 3).

A mixed-effects linear model was used to develop the bestpredictive equations for each indirect technique once it wasdetermined that the data fit the assumptions for parametricstatistics. Study site was included as a random effect in themodel, and the R2 and PRESS statistic were calculated usingthe fixed-effects equation. Residual analysis was performedto check model assumptions. To reduce the total number ofpossible equations, we selected only one canopy metric,CBDmax (maximum CBD across all 1-m canopy layers, Ta-ble 3), because it best represented the CBD needed to modelcrown fire behavior (Scott and Reinhardt 2001, 2002; vanWagner 1977). We also chose to compute CBDmax usingonly one canopy biomass aggregation: sum of all foliage andlive branchwood less than 3 mm and dead branchwood lessthan 6 mm (CBDmax2, Table 3). This biomass aggregationbest represented the fuels that would be consumed in most

© 2005 NRC Canada

730 Can. J. For. Res. Vol. 35, 2005

Variable Description

Canopy metricsCL Canopy loading (kg·m–2)CBDave Average canopy bulk density over all canopy layers (kg·m–3)CBDmaxa Canopy bulk density of canopy layer with greatest bulk density (kg·m–3)

Canopy biomass aggregations1 Canopy biomass is represented by only dead and live foliage2a Canopy biomass is defined as dead and live foliage plus live branchwood less than 3 mm in diameter and dead branch-

wood less than 6 mm in diameter; this was considered the best representation of combustible canopy fuel for thisstudy

3 Canopy biomass is all branch and foliage biomass but no stems4 Canopy biomass includes all foliage, branches, and stems

Indirect techniquesLAI2000 LI-COR LAI-2000 leaf canopy analyzerHemiphoto Hemispherical photographyCIDpci CID digital plant canopy imagerCeptometer AccuPAR ceptometerDensiometer Spherical densiometerPoint Vertical point sampling technique

Indirect analysis schemesA Average gap fraction across all five zenith anglesB Average gap fraction across top three zenith anglesC Gap fraction for top zenith angle (0°–15°)D Average gap fraction across five zenith angles using a transformation of eq. 3: CBD = ∑ −(ln ) sin cos /Ti i iθ θ 2E Average gap fraction across top three zenith angles using a transformation of eq. 3: CBD = ∑ −(ln ) sin cos /Ti i iθ θ 2

Note: Canopy metrics and biomass aggregations represent the 12 canopy characteristic expressions (represented by canopy metric plus definition suffix;for example, CL1 is canopy loading computed from only foliage) used to evaluate indirect technique performance. Indirect measurement techniques andanalysis schemes represent the 19 indirect measurement methods employed in this study (indirect schemes were not used with ceptometer, densiometer,point, and allometric methods; that is, LAI2000A is gap fraction averaged across five zenith angles for LI-COR LAI-2000 instrument).

aIndicates the canopy biomass aggregation selected to best represent canopy fuels for fire modeling in the regression analysis.

Table 3. Factors employed in the evaluation of indirect techniques to estimate canopy characteristics.

crown fires and includes the fuels used in most fire behaviormodels (Sando and Wick 1972; Scott and Reinhardt 2002).Other stand characteristic variables (stand height, basal area,and measured SLA) were included in the regression equa-tion if they significantly (p < 0.05) improved the model. Wealso developed CBD regression equations using measuredLAI from the LAI2000, hemiphoto, and CIDpci to CBDmax,because this LAI is an easily obtained measurement for theseinstruments.

Results

Profiles of canopy bulk density were computed from thedeveloped biomass equations for each study site for each ofthe canopy biomass aggregations (Fig. 2). Maximum CBDsand stand heights vary across each site because of differencesin stand height, species composition, and site productivity.Tree stems and large branches account for the majority ofCBD estimates, especially at lower levels in the canopy. Thispresents a dilemma because foliage and small branches areconsidered the primary fuels for crown fires, but the gapfraction estimated by all the indirect techniques includes allcanopy materials. The highest canopy bulk densities of com-bustible biomass are found in the mid-canopy; however, ifall crown components are summed, the highest bulk densi-ties are at the canopy base, where tree stems are found(Fig. 2). The greatest amount of combustible canopy bio-mass, and as a result, the densest combustible canopy layers,are contained in the overstory canopy layers occupied by thelargest trees. The removal of smaller trees in the initial har-vest treatments did little to reduce CBD in these overstorylayers (Scott and Reinhardt 2002).

Indirect method evaluationResults of the correlation analysis for all combinations

of canopy metrics and biomass aggregations with indirectmeasurements of gap fraction are summarized in Fig. 3. Ingeneral, correlation coefficients were above 0.50 and signifi-cant, but it was difficult to determine the best indirect tech-nique across the canopy representations because of greatvariation in measurements and the low number of observa-tions (n = 23). The ceptometer, densiometer, and point sam-pling techniques (most correlation coefficients < 0.6) did notperform as well as the LAI2000, hemiphoto, and CIDpci(correlation coefficients > 0.6). The LAI2000, hemiphoto,and CIDpci consistently perform well across all canopy bio-mass aggregations and for all canopy metrics of CBD. TheCIDpci appeared to perform the best with the exception offoliage only (Fig. 3a) and all crown material (Fig. 3d) bio-mass aggregations. The correlation coefficients for canopyloading (CL) were relatively weak with all indirect tech-niques except hemispherical photography, because the CLmeasurement does not incorporate the vertical distribution ofcanopy material across the stand profile. The ceptometer,densiometer, and point techniques seemed best able to pre-dict CL, especially when foliage and all branchwood bio-mass components are aggregated (Fig. 3c).

As expected, canopy biomass aggregations that includemost crown components — foliage and all branches (Fig. 3c)and all canopy components including stems (Fig. 3d) —correlated best with almost all indirect techniques. This is

© 2005 NRC Canada

Keane et al. 731

Fig. 2. Vertical profiles of measured canopy bulk density(CBD, kg·m–3) for each canopy biomass aggregation (see Table 3for details) for each study site: (a) Ninemile, (b) Salmon,(c) Flagstaff, (d) Blodgett, and (e) Tenderfoot. Note the heightscales are different for Ninemile and Blodgett to fully portrayCBD height distributions.

because readings from the indirect instruments include allcrown material in the computation of gap fraction, includingstems and large branches. However, it was interesting thatthe differences between biomass aggregations were relativelysmall across all canopy metrics except for canopy loading(CL) (Fig. 3). This might indicate that the projection of fo-liage can cover most supportive tissue and even some largestems (Fassnacht et al. 1994; Stenberg et al. 1994). Most in-

direct techniques had the best correlation coefficients whenCBDmax was used for foliage and small branches, exceptfor hemiphoto, which seemed to correlate best with CBDavefor all but the total biomass aggregation.

The best indirect analysis schemes for estimating canopycharacteristics (CL, CBDave, CBDmax) for the hemiphoto,LAI2000, and CIDpci techniques appear to be the schemesthat use only the top zenith angle and the ones that correctgap fraction by light-beam length across the three or five ze-nith angle categories (schemes C, E in Table 3) (Fig. 4).Correlations for the top zenith angle are high for most tech-niques because there are a large number of grid measure-ments (n = 25) within the small study site plot. Canopy fuelsdefined by only foliage and branchwood and by all crownmaterial (shown in Figs. 3c, 3d) appear to have the highestcorrelation with all three instruments and all schemes. Again,canopy loading, CL, has the poorest performance, especiallyfor the LAI2000 and CIDpci, because the measure cannotdistinguish canopy depth. The CBDmax is the canopy metricthat consistently correlates best with the LAI2000 for all fiveschemes. It also happens that the weighted average of thetop three or five zenith angles correlates best with most can-opy biomass fuel representations (Fig. 4). Correlations be-tween canopy biomass aggregations across all schemes werequite high (most r > 0.9), meaning canopy fuel aggregationshave little influence on the scheme used to compute gapfraction for the three indirect methods.

We found that the indirect methods had a slightly betterrelationship to the CBDmax canopy metric than to the CBDaveand CL (Fig. 3) and it best characterizes canopy fuel avail-able for crown fires. This supports our selection of CBDmaxas the only canopy measure to use in the regression analysis.Because there are few differences between correlation coef-ficients across all canopy biomass aggregations, there arefew negative consequences in the selection of only the fo-liage and small branch aggregation for computing CBDmaxacross all indirect techniques.

Increasing the number of sample grid points did not im-prove the estimate of canopy bulk density (CBDmax2) usingthe LAI2000, hemiphoto, and CIDpci indirect techniques(Fig. 5). The highest correlations were achieved when all 25grid points were used to compute average gap fraction. Cor-relation coefficients did not significantly change when onlynine points were used, but they were highly variable and in-consistent when only the center grid point was used.

Predictive CBD equationsRegression equations for predicting CBD (CBDmax2) from

gap fraction and stand variables performed well for theLAI2000, CIDpci, and hemiphoto (R2 > 0.60) and poorly forthe ceptometer, densiometer, and point (R2 < 0.50) (Table 4).Two or three regression equations are shown for each indi-rect technique; one equation uses the gap fraction estimatedby the indirect equipment to estimate CBD, while the otherequations includes any gap fraction transformation and standvariable that may improve the relationship. The relationshipof the CBDmax2 to gap fraction measured from the six indi-rect techniques over all five study sites is graphed in Fig. 6.These equations can be used to compute CBD from gapfraction measurements from any of the indirect techniques inthe field.

© 2005 NRC Canada

732 Can. J. For. Res. Vol. 35, 2005

Fig. 3. Absolute values of correlation coefficients for gap frac-tions measured from the six indirect techniques with the threecanopy metrics (CL, CBDave, and CBDmax; see Table 3 for def-initions). Each graph represents a different aggregation of thecanopy biomass components used to compute the canopy mea-sures: (a) foliage only, (b) foliage and combustible branchwood,(c) foliage and all branchwood, and (d) all crown fuels (foliage,branchwood, and stems). Gap fractions for LAI2000, hemiphoto,and CIDpci were calculated as a weighted average of the topthree zenith angles. Line graphs were used to best portray thedifferences between canopy metrics and do not signify a continu-ous relationship between techniques.

Gap fraction seems to be linearly related to CBDmax2(Fig. 6), and the inclusion of stand variables in the regres-sion equations improved the performance across all indirecttechniques (Table 4). Predictions of CBDmax2 from the in-direct LAI estimate computed from the LAI2000, CIDpci,and hemiphoto did surprisingly well (R2 > 0.60, Table 4). In-terestingly, regression equations with stand variables onlyalso did well for predicting CBDmax2 (Table 4), and thevariables SLA (specific leaf area) and SBA (stand basal area)significantly improved the regressions for the hemisphericalphotography and densiometer. Stand height (HEIGHT) im-proved predictions for the CIDpci (Table 4).

Discussion

The relationship of gap fraction to CBD is greatly influ-enced by vegetation and site conditions on the study plot,presumably because of the differences in species and standmorphological characteristics. These compositional and struc-tural differences are responsible for the high variation shown

© 2005 NRC Canada

Keane et al. 733

Fig. 4. Absolute values of correlation coefficients for gap frac-tions measured from the LAI2000, hemiphoto, and CIDpci meth-ods for all canopy metrics (CL, CBDave, and CBDmax; seeTable 3) but for only one definition of canopy fuels (foliage andsmall branches). Each graph represents a different indirectscheme to estimate gap fraction: (a) average of five zenith angleranges, (b) average of three zenith angles, (c) top zenith angle,(d) weighted average of top five zenith angles, and (e) weightedaverage of top three zenith angle ranges (see Table 3 for moreinformation). Line graphs were used to best portray the differ-ences between canopy metrics and do not signify a continuousrelationship between techniques.

Fig. 5. Relationship of canopy bulk density with gap fraction.CBD was computed as a maximum across all 1-m canopy layersusing the foliage and small branchwood biomass aggregation(CBDmax2) with gap fraction measured from the six indirect meth-ods: (a) LAI2000, (b) hemiphoto, and (c) CIDpci, using resultsfrom the top three zenith angles (see Table 3 for more information).

in Fig. 6. The R2 increases by 0.10–0.20 if stand variablesare included in the regression analysis (Table 4), probablybecause these variables may minimize the subtle differencesbetween study sites. Major stand variables that differ greatly

across study sites include stand height, species composition,and stand density. Stand height would dictate the distancethat light must travel through the canopy to be detected bythe instrument (see eq. 2). This is especially important when

© 2005 NRC Canada

734 Can. J. For. Res. Vol. 35, 2005

Equation Regression coefficients (SE) p value R2 PRESS

LI-COR LAI-2000 (LAI2000)� = 0.2231 – 0.2012 LAI2000A Intercept: 0.2231 (0.0226) <0.0001 0.558 0.0387

LAI2000A: –0.2012 (.0245) <0.0001� = 0.0402 + 7.6293 LAI2000E Intercept: 0.0402 (0.0167) 0.0286 0.696 0.0269

LAI2000E: 7.6293 (0.7713) <0.0001Hemispherical photography (Hemiphoto)� = 0.2147 – 0.2726 HemiphotoA Intercept: 0.2147 (0.0272) <0.0001 0.359 0.0579

HemiphotoA: –0.2726 (0.0375) <0.0001� = 0.028 + 2.257 HemiphotoE Intercept: 0.028 (0.0197) 0.1704 0.629 0.0373

HemiphotoE: 2.257 (0.2988) <0.0001� = 0.299 + 1.166 HemiphotoE – 0.022 SLA – 0.003 SBA +

0.00028 SLA×SBAIntercept: 0.299 (0.081) 0.0027 0.687 0.0162

HemiphotoE: 1.166 (0.6530) 0.0976SLA: –0.022 (0.0065) 0.0044SBA: –0.003 (0.0005) <0.0001SLA×SBA: 0.00028 (0.00005) <0.0001

CID plant canopy imager (CIDpci)� = 0.4777 – 0.6119 CIDpciA Intercept: 0.4777 (0.0655) <0.0001 0.515 0.0385

CIDpciA: –0.6119 (0.1018) 0.0001� = 1.5579 – 3.6370 CIDpciA + 2.4666 (CIDpciA)2 – 0.00681

HEIGHTIntercept: 1.5579 (0.234) 0.0001 0.944 0.0071

CIDpciA: –3.6370 (0.757) 0.0010(CIDpciA)2: 2.4666 (0.617) 0.0031Height: 0.00681 (0.0074) 0.0027

AccuPAR ceptometer (ceptometer)� = 0.1802 – 0.1284 ceptometer Intercept: 0.1802 (0.0280) <0.0001 0.187 0.0751

Ceptometer: –0.1284 (0.0228) <0.0001� = 0.1044 – 0.0730 ceptometer + 0.00036 SBA Intercept: 0.1044 (0.0372) 0.0127 0.314 0.0575

Ceptometer: –0.0730 (0.0273) 0.0166SBA: 0.00036 (0.000125) 0.0109

Stand densiometer (densiometer)� = 0.1457 – 0.1353 densiometer Intercept: 0.1457 (0.0279) 0.0001 0.043 0.0959

Densiometer: –0.1353 (0.0404) 0.0038Point sampling (point)� = 0.1884 – 0.1562 point Intercept: 0.1884 (0.0308) <0.0001 0.192 0.0747

Point: –0.1562 (0.0402) 0.0012Stand-related variables only� = 0.295 – 0.0219 SLA – 0.00319 SBA + 0.00032 SLA×SBA Intercept: 0.295 (0.0769) 0.0016 0.633 0.0284

SLA: –0.0219 (0.0061) 0.0028SBA: –0.00319 (0.00048) <0.0001SBA×SLA: 0.00032 (0.000040) <0.0001

Indirect measurement of LAI� = 0.0424 + 0.06778 LAILAI2000 Intercept: 0.0424 (0.0178) 0.0300 0.642 0.0321

LAI: 0.06778 (0.0073) <0.0001� = 0.0396 + 0.0511 LAIHemiphoto Intercept: 0.0396 (0.0293) 0.1945 0.616 0.0795

LAI: 0.0511 (0.0094) 0.0001� = –0.0234 + 0.2056 LAICIDpci Intercept: –0.0234 (0.0203) 0.2772 0.784 0.0169

LAI: 0.2056 (0.0236) <0.0001

Note: The term � denotes CBDmax (kg·m–3). The suffix A in each indirect technique label refers to the average gap fraction across all zenith angles,and the suffix E refers to the weighted average gap fraction across all three top angles (Table 3). SLA refers to the average specific leaf area (m2·kg–1) ofthe canopy, SBA refers to stand basal area (m2·ha–1), and HEIGHT is the average stand height (m). Also shown are regression equations for predictingCBDmax from stand variables only and from LAI as estimated by the indirect technique.

Table 4. Results of regression analysis to predict the maximum canopy bulk density (CBDmax2) of the foliage and small branchwoodbiomass aggregation for all indirect techniques.

light is coming from low zenith angles. Species composi-tion, coarsely represented by SLA in our analysis, dictatesfoliage morphological characteristics, such as clumping andneedle length, which influences the probability that lightwill hit a foliar particle. Stand density also influences thechance that stems or large branches will intercept light. Themost precise CBD regression equations include these stand-related factors.

The performance of most indirect techniques to predictCBD was lowered when combustible canopy biomass (fo-liage and small branchwood) was used instead of foliageonly or all canopy biomass material. Combustible biomass isrequired by most crown fire models, so regression equationswere constructed with this canopy definition, but the precisionof the CBD estimate using these biomass components waslower. Fortunately, the best CBD estimates were obtainedwhen CBDmax was used, which is probably the best CBDmeasure for crown fire modeling (Scott and Reinhardt 2002).

Evaluation resultsThe LAI2000, CIDpci, and hemispherical photography were

the best techniques for predicting crown fuel characteristics,especially CBDmax. Even though the predictive equationsappear inexact (0.60 < R2 < 0.95, Table 4), we feel they areadequate for estimating CBDmax, given the predictive capa-bility of the fire models. However, each instrument has someadvantages and disadvantages for field use. The LAI2000calculates gap fraction from differences in light intensity, andtherefore its accuracy is greatly determined by the variabilityof light conditions outside and below the canopy during sam-pling. Frequently changing light conditions, such as those ex-perienced on windy or partly cloudy days, can greatly influencethe computation of gap fraction; it is best used at dawn ordusk on clear or fully cloudy days (Welles and Norman1991; LI-COR 1992). Furthermore, the measured light inten-sity data are for only five static zenith angles, whereas hemi-spherical photography can produce an image that can beanalyzed with any number of zenith angles over a full 360°horizontal view.

The CIDpci performance was quite similar, if not superior,to hemispherical photography because the instruments areessentially the same, except that the CIDpci camera is em-bedded in an easily used wand, the image resolution is coarser,and the analysis software uses different analysis algorithms.The CIDpci and hemispherical photography performedequivalently (Fig. 3); however, hemispherical photography isless expensive (Table 1). Hemispherical photography can besomewhat difficult to obtain, especially if multiple points aredesired to describe variability of stand conditions, becausethe camera must be leveled on a tripod and the correct expo-sure and shutter speed must be determined. The LAI2000measurements may be easier to obtain in the field becausethe wand can be readily positioned in the appropriate posi-tion and a reading taken quickly. Hemiphoto and CIDpci donot require calibration measurements (outside canopy) sothey can be used in stands where large calibration openingsare distant or rare; however, both are sensitive to the effectsof diffuse and direct radiation. Equipment for hemisphericalphotography costs less than the LAI2000 and CIDpci, and itcan be used for many other purposes (Table 1). Additionalfield crew training may be required for the LAI2000.

© 2005 NRC Canada

Keane et al. 735

Fig. 6. Canopy bulk density computed as a maximum across all1-m canopy layers using the foliage and small branchwood bio-mass aggregation (CBDmax2) with gap fraction measuredfrom the six indirect methods: (a) LAI2000, (b) hemiphoto,(c) CIDpci, (d) ceptometer, (e) densiometer, and ( f ) point,using results from the best analysis scheme (see Table 3 formore information).

The remaining indirect techniques did not perform as wellas the LAI2000, hemiphoto, and CIDpci, but they might alsohave value for measuring CBD under some circumstances.The poor performance of the ceptometer (Fig. 3) is mainlydue to the sensors receiving light from all canopy zenith an-gles, so corrections to light path length are impossible. Dy-namic sky conditions also affected ceptometer accuracy. Itshigh cost (Table 1) and its calibration measurement require-ment, coupled with its low performance, make it a poorchoice for CBD measurements, but others have found it workedwell on sunny days (White et al. 1998). The densiometer andpoint sampling methods provide a low-cost alternative toCBD measurement, but the low performance coupled withthe high number of sample points needed to obtain an accurateestimation of gap fraction probably limits their application.

The allometric approach is a viable alternative to the indirectapproaches based on gap fraction presented here, providingthere are accurate biomass equations for local applications(Reinhardt and Crookston 2003). The allometric techniquerequires an extensive inventory of all trees on the plot in-cluding seedlings and saplings, where the height, crown baseheight, and diameter at breast height are measured for eachtree. This inventory can take much longer than it takes tomeasure gap fraction with an optical sensor, but the data canbe used for many other forestry applications. Moreover, theabundant forest inventory data previously taken by land man-agement institutions and agencies can be used to computeCBD across large land areas (Keane et al. 2000). It is diffi-cult and somewhat inappropriate to statistically compare CBDestimates from the allometric technique with those measuredfrom these other indirect techniques. The allometric tech-nique depends on accurate biomass equations and generalassumptions about crown shape and random tree distributionto determine local crown fuel characteristics.

Scale analysisThe low variation of indirect measurements across the 25

grid points (Fig. 5) is primarily a scale artifact of our sam-pling protocol rather than a comprehensive assessment ofgap fraction spatial variation. We intensively sampled crownbiomass using destructive methods in a small circular 0.3-harepresentative portion of the stand instead of sampling theentire stand because crown biomass sampling is costly andtime consuming. As a result, our indirect measurements hadto be taken on a 5-m grid confined to this small sampling areato ensure the indirect measurements reflected the destructivelysampled canopy fuel conditions. The 25 grid sample pointsare so close together that the indirect instruments often mea-sure the same canopy conditions, because the area scannedby the indirect equipment (field of view) was much greaterthan the area occupied by the sample plot area. For all studysites, it appears that only three to five measurements wereneeded to obtain an accurate gap fraction for all indirecttechniques to measure the small plot area because of the sen-sor’s large measurement footprint.

It is important that measurement and analysis scales matchwhen estimating CBD from indirect techniques. If CBD esti-mates are needed at the plot level, then only three to five in-direct measurements may be needed, but it can be useful totake additional readings, especially if canopy conditions in

plot are highly variable. However, if one CBD estimate isneeded for an entire stand or polygon, then the number ofindirect sampling points should be sufficient to capture theheterogeneity of canopy conditions across the entire extentof the stand. A good guide might be to use standard sam-pling designs employed in forestry for tree inventory to de-termine the number of CBD indirect sample points.

Limitations of the indirect approachAlthough results from this study are promising, there are

major limitations in measuring canopy bulk density usingindirect techniques. First, all indirect methods are greatlyinfluenced by leaf and branch morphology and geometry(Bolstad and Gower 1990; Stenberg 1996a), which is themain reason that the relationship of CBD with gap fractionis different across sites of dissimilar species compositionand vertical structures (Fig. 6). All estimations of gap frac-tion assume a random distribution of material throughout thecanopy, which is not the case for most stands. Foliage inpine species, for example, tends to be clustered at the end ofthe branch, while fir foliage is well distributed along most ofthe branch length. The specific leaf area (SLA) used in re-gression analysis was not an important predictive variable(Table 4) because it was calculated by species at the branchlevel and was not adjusted for branch clumping on the tree(Stenberg 1996b). Moreover, SLA can vary throughout thecrown profile within a tree species. Measuring specific pro-jected crown area instead of specific leaf area, where needleand branch mass is divided by the area the branch projectson the ground at the tree level, might improve the predic-tions.

Light has a greater chance of being intercepted by canopymaterial if detected from high zenith angles (not directlyabove), because it must traverse a greater distance throughthe canopy. Equation 1 attempts to correct for this by weight-ing by the sine and cosine function, but this also assumesthat the light is passing through a homogeneous medium,which is not always the case in forests. The sample standmight be on a rounded ridge, steep slope, or deep valley,where the light path through the canopy is either reduced ortruncated by topography or some other abiotic factor. Alter-natively, highly variable crown conditions within the standdue to clumping, shading, sun angle, or dead aerial material(needle drape or lichen) may influence accurate estimationof CBD, depending on the sampling strategy. Because notested instrument can detect canopy depth, indirect measure-ments estimate gap fraction across the entire canopy profile,so dense canopy layers cannot be directly measured for CBD.

Canopy fuel characteristics are highly variable in spaceand time, making it difficult to consistently estimate CBDwith the optical indirect techniques presented here. Not onlyis there a large horizontal variation at small spatial scalesdue to tree distributions, but the vertical distribution of crownfuels is also highly variable because of diverse distributionsof tree sizes and shapes. Sampling of this variability cantake numerous measurements over large areas, which maynot be feasible for many fuel sampling projects. This limita-tion is most visible in the regression analysis results in Fig. 6,where most variation is due to across-stand differences inspecies composition and structure. In fact, gap fraction pre-

© 2005 NRC Canada

736 Can. J. For. Res. Vol. 35, 2005

dictions of CBDmax2 had R2 > 0.90 for many optical meth-ods when the regression analysis was stratified by study site(n = 4). Regression equations presented in Table 4 will per-form best in stands of similar species composition and struc-ture as those used in this study (Table 2).

Tree, branch, and foliar clumping also contribute to can-opy fuel variability, and these factors are responsible formost errors in predicting CBD (Fig. 6). Smith et al. (1993)found that the nonrandom distribution of branches in spacecaused low LAI2000 LAI measurements, and a clumpingfactor of 2.63 was used to correct LAI2000 estimates tothose predicted from allometric equations. Sampson and Allen(1995) found a weak relationship of LAI measurements fromthe LAI2000 to allometric equations or litter-trap estimatesat very low and high LAIs, which was probably due to crownspatial structure. Strachan and McCaughey (1996) found that84 LAI2000 measurements are needed to get within 5% ofthe observed mean for a stand whose crown was highly vari-able, but only 21 measurements are needed to get to 10% ofthe mean.

Measurement error can be a major factor in the accuracyof gap fraction measurement. Improper calibration measure-ments (LAI2000, ceptometer) or camera settings (hemiphoto,CIDpci) can result in inaccurate CBD estimations. The in-strument should be positioned to ensure topography, rocks,and people are not in the image frame to achieve the bestgap fraction estimates. Measurements for all indirect tech-niques must be taken when ambient light conditions do notchange over short periods. These operator errors are proba-bly the primary explanations for the obvious outliers in Fig. 6for the Salmon LAI2000 and hemiphoto measurements atlow gap fractions.

Our study had only 23 observations to build predictivemodels for indirect measurement techniques. This is proba-bly is not enough to adequately sample the full range of can-opy conditions. Moreover, there were only five study siteswith 18 of the total 23 observations created by tree removal,which may contribute to some bias in the sample. Additionalmeasurements across disparate sites and species composi-tions are needed to build stronger predictive models for indirecttechniques. Moreover, the 23 observations are not entirelyindependent because some trees are present in all replicatemeasurements. We recommend these procedures be repeatedin other conifer and broadleaf stands to develop more robustrelationships between canopy characteristics and gap frac-tion.

The study can be of immediate benefit to managers andresearchers by providing a means to estimate canopy bulkdensity in the field using the regression equations presentedhere with gap fraction measurements from the indirect tech-niques. These field measurements can be used to drive clas-sifications of satellite imagery to map CBD across a spatialdomain (see Keane et al. 2001) and input this digital map inspatial fire behavior prediction systems such as FARSITE(Finney 1998b). Stand-level measurements of CBD can beused in nonspatial models such as NEXUS to calculate crownfire characteristics (Scott 1999). To properly use these re-gression equations, the intermediate data calculated by theindirect instrument are required to get the gap fraction forthe five zenith angle ranges. Then, the estimates of gap frac-

tion are transformed using eq. 3 or Table 3 for the appropriatenumber of zenith angle ranges.

Acknowledgements

This work was funded by a grant from the USDA and theInterior Joint Fire Science Program. It was also supported bythe USDA Forest Service Rocky Mountain Research Station(RMRS) and Systems for Environmental Management. Wethank Steve Slaughter and Laura Ward, Ninemile RangerDistrict, Lolo National Forest; Terry Hershey and BarbaraLevesque, Salmon-Cobalt Ranger District, Salmon-ChallisNational Forest; Allen Farnsworth and Chuck McHugh,Coconino National Forest; Bob Heald, Jason Moghaddas,Frieder Schurr, and Sheryl Rambeau, University of Califor-nia Center for Forestry, Blodgett Forest Research Station;and Ward McCaughey and Leon Theroux, RMRS MissoulaForestry Sciences Laboratory for facilitating this work. Wethank the canopy field crew: Kylie Kramer, MatthewDuveneck, Dustin Walters, Bill Ballinger, Niki Parenteau,Courtney Couch, Cassie Koerner, Kate Dirksen, AndrewChristie, and Roham Abtahi. We also thank reviewers JosephWhite of Baylor University, Matthew Duvenek of the Uni-versity of Massachusetts, and Mike Keim of the Universityof Washington. The use of trade or firm names in this paperis for reader information and does not imply endorsement bythe USDA of any product or service. The US Government isauthorized to produce and distribute this work for govern-ment purposes notwithstanding any copyright notation thatmay appear hereon.

References

Alemdag, I.S. 1980. Manual of data collection and processing forthe development of forest biomass relationships. Can. For. Serv.Petawawa Natl. For. Inst. Inf. Rep. PI-X-4.

Alexander, M.E. 1988. Help with making crown fire hazard assess-ments. In Protecting People and Homes from Wildfire in theInterior West: Proceedings of the symposium and workshop.Edited by W. Fischer and S.F. Arno. USDA For. Serv. Gen.Tech. Rep. INT-251. pp. 147–153.

Arno, S.F., Parsons, D.J., and Keane, R.E. 2000. Mixed-severityfire regimes in the northern Rocky Mountains: consequences offire exclusion and options for the future. In Wilderness Sciencein a Time of Change Conference. Vol. 5: Wilderness Ecosys-tems, Threat, and Management, Missoula, Montana, 23–27 May1999. USDA For. Serv. RMRS-P-15-VOL-5. pp. 225–232.

Battles, J.J., Dushoff, J.G., and Fahey, T.J. 1996. Line intersectsampling of forest canopy gaps. For. Sci. 42: 131–140.

Bolstad, P.V., and Gower, S.T. 1990. Estimation of leaf area indexin fourteen southern Wisconsin forest stands using a portable ra-diometer. Tree Physiol. 7: 115–124.

Brown, J.K. 1978. Weight and density of crowns of Rocky Moun-tain conifers. USDA For. Serv. Res. Pap. INT-197.

Campbell, G.S., and Norman, J.M. 1989. The description and mea-surement of plant canopy structure. In Plant canopies: theirgrowth, form, and function. Edited by P.G. Jarvis. CambridgeUniversity Press, Cambridge, UK. pp. 1–19.

Chason, J.W., Baldocchi, D.D., and Huston, M.A. 1991. A compar-ison of direct and indirect methods for estimating forest canopyleaf area. Agric. For. Meteorol. 57: 107–128.

© 2005 NRC Canada

Keane et al. 737

Chen, S.G., Impens, I., Ceulemans, R., and Kockelbergh, F. 1993.Measurement of gap fraction of fractal generated canopies usingdigitalized image analysis. Agric. For. Meteorol. 65: 245–259.

Clark, P.E., and Seyfried, M.S. 2001. Point sampling for leaf areaindex in sagebrush steppe communities. J. Range Manage. 54:589–594.

Decagon. 1987. Sunfleck ceptometer user’s manual. Decagon De-vices, Pullman, Wash.

Deusen, P.C.V., and Baldwin, V.C. 1993. Sampling and predictingtree dry weight. Can. J. For. Res. 23: 1826–1829.

Dufrene, E., and Breda, N. 1995. Estimation of deciduous forestleaf area index using direct and indirect methods. Oecologia,104: 156–162.

Englund, S.R., O’Brien, J.J., and Clark, D.B. 2000. Evaluation ofdigital and film hemispherical photography and sphericaldensiometry for measuring forest light environments. Can. J.For. Res. 30: 1999–2005.

Fassnacht, K.S., Gower, S.T., Norman, J.M., and McMurtie, R.E.1994. A comparison of optical and direct methods for estimatingfoliage surface area index in forests. Agric. For. Meteorol. 71:183–207.

Ferry, G.W., Clark, R.G., Montgomery, R.E., Mutch, R.W.,Leenhouts, W.P., and Zimmerman, G.T. 1995. Altered fireregimes within fire-adapted ecosystems. US Department of theInterior, National Biological Service, Washington, D.C.

Finney, M.A. 1998. FARSITE users guide and technical documen-tation. USDA For. Serv. Res. Pap. RMRS-RP-4.

Frazer, G.W., Trofymow, J.A., and Lertzman, K.P. 1997. A methodfor estimating canopy openness, effective leaf area index, andphotosynthetically active photo flux density using hemisphericalphotography and computerized image analysis techniques. Can.For. Serv. Pac. For. Cent. Inf. Rep. BC-X-373.

Frazer, G.W., Canham, C.D., and Lertzman, K.P. 2000. Technolog-ical tools: Gap Light Analyzer version 2.0. Bull. Ecol. Soc. Am.12: 191–197.

Frazer, G.W., Fournier, R.A., Trofymow, J.A., and Hall, R.J. 2001.A comparison of digital and film fisheye photography for analy-sis of forest canopy structure and gap light transmission. Agric.For. Meteorol. 109: 249–263.

Gary, H.L. 1976. Crown structure and distribution of biomass in alodgepole pine stand. USDA For. Serv. Res. Pap. RM-165.

Gary, H.L. 1978. The vertical distribution of needles and branch-wood in thinned and unthinned 80-year-old lodgepole pine.Northwest Sci. 52: 303–309.

Gower, S.T., and Norman, J.M. 1991. Rapid estimation of leaf areaindex in conifer and broad-leaf plantations. Ecology, 72: 1896–1900.

Groeneveld, D.P. 1997. Vertical point quadrat sampling and an ex-tinction factor to calculate leaf area index. J. Arid Environ. 36:475–485.

Inouye, D.W. 2000. Gap Light Analyzer version 2.0. Bull. Ecol.Soc. Am. 6: 191–197.

Johnson, A.F., Woodard, P.M., and Titus, S.J. 1989. Lodgepolepine and white spruce crown fuel weights predicted from heightand crown width. Can. J. For. Res. 19: 527–530.

Jurik, T.W., Briggs, G.M., and Gates, D.M. 1986. A comparison offour methods for determining leaf area index in successionalhardwood forests. Can. J. For. Res. 15: 1154–1158.

Keane, R.E., Garner, J.L., Schmidt, K.M., Long, D.G., Menakis,J.P., and Finney, M.A. 1998. Development of input spatial datalayers for the FARSITE fire growth model for the Selway–Bitterroot Wilderness complex, USA. USDA For. Serv. Gen.Tech. Rep. RMRS-GTR-3.

Keane, R.E., Mincemoyer, S.A., Schmidt, K.M., Menakis, J.P.,Long, D.G., and Garner, J.L. 2000. Mapping vegetation and fu-els for fire management on the Gila National Forest Complex.USDA For. Serv. Gen. Tech. Rep. RMRS-GTR-46-CD.

Keane, R.E., Burgan, R.E., and Wagtendonk, J.V. 2001. Mappingwildland fuels for fire management across multiple scales: inte-grating remote sensing, GIS, and biophysical modeling. Int. J.Wildland Fire, 10: 301–319.

Keane, R.E., Veblen, T., Ryan, K.C., Logan, J., Allen, C., andHawkes, B. 2002. The cascading effects of fire exclusion in theRocky Mountains. In Rocky Mountain futures: an ecologicalperspective. Edited by J.S. Baron. Island Press, Washington,D.C. pp. 133–153.

Kolb, P.F., Adams, D.L., and McDonald, G.I. 1998. Impacts of fireexclusion on forest dynamics and processes in central Idaho.Tall Timbers Fire Ecol. Conf. 20: 911–923.

Kucharik, C.J., Norman, J.M., and Gower, S.T. 1998. Measure-ments of leaf orientation, light distribution, and sunlit leaf areain a boreal aspen forest. Agric. For. Meteorol. 91: 127–148.

LI-COR, Inc. 1992. LAI2000 Plant Canopy Analyzer: operatingmanual. LI-COR Inc., Lincoln, Nebr.

Linn, R.R. 1997. A transport model for prediction of wildfire behav-ior. Ph.D. thesis, New Mexico State University, Las Cruces, N.M.

Magnussen, S., and Boudewyn, P. 1998. Derivations of stand heightsfrom airborne laser scanner data with canopy-based quartile esti-mators. Can. J. For. Res. 28(7): 1016–1031.

Martens, S.N., Ustin, S.L., and Rousseau, R.A. 1993. Estimation oftree canopy leaf area index by gap fraction analysis. For. Ecol.Manage. 61: 91–108.

Miller, J.B. 1967. A formula for average foliage density. Aust. J.Bot. 15: 141–144.

Monserud, R.A., and Marshall, J.D. 1999. Allometric crown rela-tions in three norther Idaho conifer species. For. Sci. 29: 521–535.

Mutch, R.W. 1994. Fighting fire with fire: a return to ecosystemhealth. J. For. 92: 31–33.

Nel, E.M., and Wessman, C.A. 1993. Canopy transmittance modelsfor estimating forest leaf area index. Can. J. For. Res. 23: 2579–2586.

Neumann, H.H., Hartog, G.D., and Shaw, R.H. 1989. Leaf areameasurements based on hemispheric photographs and leaf-littercollection in a deciduous forest during autumn leaf-fall. Agric.For. Meteorol. 45: 325–345.

Nilson, T. 1999. Inversion of gap frequency data in forest stands.Agric. For. Meteorol. 98–99: 437–448.

Nilson, T., Anniste, J., Lang, M., and Praks, J. 1999. Determina-tion of needle area indices of coniferous forest canopies in theNOPEX region by ground-based optical measurements and sat-ellite images. Agric. For. Meteorol. 98–99: 449–462.

Peper, P.J., and McPherson, E.G. 1998. Comparison of five meth-ods for estimating leaf area index of open-grown deciduoustrees. J. Arboric. 24: 98–111.

Pierce, L.L., and Running, S.W. 1988. Rapid estimation of conifer-ous forest leaf area index using a portable integrating radiome-ter. Ecology, 69: 1762–1767.

Pierce, L.L., Running, S.W., and Walker, J. 1994. Regional-scalerelationships of leaf area index to specific leaf area and leaf ni-trogen content. Ecol. Appl. 4: 313–321.

Reinhardt, E., and Crookston, N.L. (Editors). 2003. The fire andfuels extension to the Forest Vegetation Simulator. USDA For.Serv. Gen. Tech. Rep. RMRS-GTR-166.

Roberts, S.D., and Long, J.N. 1992. Production efficiency of Abieslasiocarpa: influence of vertical distribution of leaf area. Can. J.For. Res. 22: 1230–1234.

© 2005 NRC Canada

738 Can. J. For. Res. Vol. 35, 2005

Sampson, D.A., and Allen, H.L. 1995. Direct and indirect esti-mates of LAI for lodgepole and loblolly pine stands. Trees(Berl.), 9: 119–122.

Sando, R.W., and Wick, C.H. 1972. A method of evaluating crownfuels in forest stands. USDA For. Serv. Res. Pap. NC-84.

Scott, J.H. 1999. NEXUS: A system for assessing crown fire haz-ard. Fire Manage. Notes, 59: 21–24.

Scott, J.H., and Reinhardt, E.D. 2001. Assessing crown fire poten-tial by linking models of surface and crown fire behavior. USDAFor. Serv. Res. Pap. RMRS-RP-29.

Scott, J.H., and Reinhardt, E.D. 2002. Estimating canopy fuels inconifer forests. Fire Manage. Today, 62: 45–50.

Smith, R.W., and Somers, G.L. 1991. SUNSHINE: a light environ-ment simulation system based on hemispherical photographs.USDA For. Serv. Gen. Tech. Rep. SO-267.

Smith, N.J., Chen, J.M., and Black, T.A. 1993. Effects of clumpingon estimates of stand vegetation area index using the LI-CORLAI2000. Can. J. For. Res. 23: 1940–1943.

Smolander, J., and Stenberg, P. 1996. Response of LAI2000 esti-mates to changes in plant surface area index in a Scots pinestand. Tree Physiol. 16: 345–349.

Stenberg, P. 1996a. Correcting LAI2000 estimates for the clump-ing of needles in shoots of conifers. Agric. For. Meteorol. 79: 1–8.

Stenberg, P. 1996b. Simulations of the effects of shoot structureand orientation on vertical gradients in intercepted light by coni-fer canopies. Tree Physiol. 16: 99–108.

Stenberg, P., Linder, S., Smolander, H., and Flower-Ellis, J. 1994.Performance of the LAI2000 plant canopy analyzer in estimat-ing leaf area index of some Scots pine stands. Tree Physiol. 14:981–995.

Strachan, I.B., and McCaughey, J.H. 1996. Spatial and vertical leafarea index of a deciduous forest resolved using the LAI2000plant canopy analyzer. For. Sci. 42: 176–181.

van Wagner, C.E. 1977. Conditions for the start and spread ofcrown fire. Can. J. For. Res. 7: 23–34.

Walter, J.M.N., and Himmler, G.G. 1996. Spatial heterogeneity ofa Scots pine canopy: an assessment by hemispherical photogra-phy. Can. J. For. Res. 26: 1610–1619.

Welles, J.M. 1990. Some indirect methods of estimating canopystructure. Remote Sens. Rev. 5: 31–43.

Welles, J.M., and Norman, J.M. 1991. Instrument for indirect mea-surement of canopy architecture. Agron. J. 83: 818–825.

White, J.D., Running, S.W., Nemani, R., Keane, R.E., and Ryan,K.C. 1998. Measurement and mapping of LAI in Rocky Moun-tain montane ecosystems. Can. J. For. Res. 27: 1714–1727.

© 2005 NRC Canada

Keane et al. 739